The Agent Skills Directory

[COMMAND_EXECUTION]: The provided Node.js and Bash scripts use execSync to run shell commands like find, grep, and git. The scripts interpolate file paths and directory names directly into these commands, which may allow for argument injection if the agent is manipulated into using malicious input values.
[PROMPT_INJECTION]: The skill has a risk of indirect prompt injection as it processes untrusted data.
Ingestion points: Untrusted codebase files are scanned by scripts/scan.js and content is placed in .claude/codegraph.draft.md (via SKILL.md).
Boundary markers: None are used to separate scanned content from instructions.
Capability inventory: The skill uses execSync for shell commands in scripts/scan.js.
Sanitization: No validation or sanitization is performed on the content extracted from files before the agent reads and summarizes it.

code-graph